LETTER
1038/nature20817
Weak synchronization and large-scale collective
oscillation in dense bacterial suspensions
Chong Chen1*, Song Liu1*, Xia-qing Shi2, Hugues Chaté3,4 & Yilin Wu1
Collective oscillatory behaviour is ubiquitous in nature1, having
a vital role in many biological processes from embryogenesis2
and organ development3 to pace-making in neuron networks4.
Elucidating the mechanisms that give rise to synchronization is
essential to the understanding of biological self-organization.
Collective oscillations in biological multicellular systems often arise
from long-range coupling mediated by diffusive chemicals2,5–9, by
electrochemical mechanisms4,10, or by biomechanical interaction
between cells and their physical environment11. In these examples,
the phase of some oscillatory intracellular degree of freedom is
synchronized. Here, in contrast, we report the discovery of a weak
synchronization mechanism that does not require long-range
coupling or inherent oscillation of individual cells. We find that
millions of motile cells in dense bacterial suspensions can self-
organize into highly robust collective oscillatory motion, while
individual cells move in an erratic manner, without obvious periodic
motion but with frequent, abrupt and random directional changes.
So erratic are individual trajectories that uncovering the collective
oscillations of our micrometre-sized cells requires individual
velocities to be averaged over tens or hundreds of micrometres.
On such large scales, the oscillations appear to be in phase and
the mean position of cells typically describes a regular elliptic
trajectory. We found that the phase of the oscillations is organized
into a centimetre-scale travelling wave. We present a model of noisy
self-propelled particles with strictly local interactions that accounts
faithfully for our observations, suggesting that self-organized
collective oscillatory motion results from spontaneous chiral and
rotational symmetry breaking. These findings reveal a previously
unseen type of long-range order in active matter systems (those in
which energy is spent locally to produce non-random motion)12,13.
This mechanism of collective oscillation may inspire new strategies
to control the self-organization of active matter14–18 and swarming
robots19.
Like many other flagellated bacteria, Escherichia coli cells can swarm
over the surface of solid substrates such as agar plates20, forming
densely packed colonies with a surface packing fraction21 of about 50%.
Under standard growth conditions (see Methods), E. coli cells (0.8 µm
in diameter, 2–4 µm in length) produce a 5–10-µ m-thick layer of
‘swarm fluid’22 spanning most of the agar surface, in which they swim
at a mean speed of 34 µm s−1 (Extended Data Fig. 1). The resulting
quasi-two-dimensional dense bacterial suspension can persist over
large spatial scales (centimetres) for hours. We grew about one hundred
of such colonies. Approximately 12 h after inoculation, the colony has
invaded the whole dish and the cells, observed in phase-contrast videos,
display a disordered state with collective motion at small spatial scales
(a few tens of micrometres) taking the form of transient jets and vortices
(Supplementary Video 1). This ‘bacterial’ or ‘mesoscale’ turbulence has
been described recently and the rheology of concentrated suspensions
received a lot of attention from active matter physicists23–26.
1
Department of Physics and Shenzhen Research Institute, The Chinese University of Hong Kong, Shatin, Hong Kong, China. 2Center for Soft Condensed Matter Physics and Interdisciplinary
Research, Soochow University, Suzhou 215006, China. 3Service de Physique de l’Etat Condensé, CEA, CNRS, Université Paris-Saclay, CEA-Saclay, 91191 Gif-sur-Yvette, France. 4Beijing
Computational Science Research Center, Beijing 100094, China.
*These authors contributed equally to this work.
2 1 0 | N AT U R E | VO L 5 4 2 | 9 F E B R UA RY 2 0 1 7
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
doi:10.
As the cell density increases further owing to cell multiplication
(Methods), a heretofore unnoticed phenomenon emerges. Although
no apparent change seems to occur in phase-contrast videos
(Supplementary Video 2), their velocity, measured via particle image
velocimetry (PIV) analysis (Methods) and averaged over spatial scales
greater than about 100 µm undergoes regular periodic oscillations
within the plane of the swarm-fluid film (Fig. 1c and Extended Data
Fig. 2). This collective oscillatory motion (hereafter referred to as
‘collective oscillation’) is characterized by a fluctuating but featureless,
spatially homogeneous velocity field that oscillates in time. (Note
that this is fundamentally different from the rotational modes or
vortical collective motion commonly found in biological and active
matter systems under confinement27). Once emerged, the collective
oscillation can persist for at least half an hour. The oscillation period is
steady over time and ranges from 4 s to 12 s across different colonies.
The orthogonal components of the collective velocity are well fitted
by sinusoidal functions with identical period but different amplitude
and phase (Extended Data Fig. 2), indicating that the average positions
of cells follow elliptical trajectories. We observed the same collective
oscillation in swarms of Proteus mirabilis, a species similar to E. coli in
terms of motile behaviour but distinct in biochemistry.
The collective oscillation was also clearly manifested when we visua-
lized the flow of the upper swarming fluid using silicone-oil droplets
with minimal invasiveness (Methods). Strikingly, these tracers followed
smooth and oscillatory elliptical trajectories with quasi-synchronized
phases even when they were separated by hundreds of micrometres
(Fig. 1b and Supplementary Video 2). Their velocity and the collective
velocity of cells were in synchrony as well (Fig. 1c), suggesting that the
oscillatory flow was generated by the collective motion of cells that drag
nearby fluid along28. The amplitude of the collective velocity was some-
what smaller than that of the tracers owing to the averaging procedure.
In our experiments we found elliptical collective oscillations most
often (68 out of 71 cases), with rare exceptions being linear or irregular
oscillations (3 out of 71 cases; Extended Data Fig. 3). In almost all cases,
the chirality and the long-axis orientation of collective oscillations
are the same within a specific colony, but the chirality has an equal
probability of being clockwise or counterclockwise across different
colonies (Extended Data Fig. 3a), indicative of spontaneous global chiral
symmetry breaking. The long-axis orientation across different colonies
is non-uniformly distributed (Fig. 2a), probably reflecting the aniso-
tropy of the large-scale geometry of colonies (Extended Data Fig. 1).
These results imply that the collective oscillation of cells is correlated
over very long distances. To verify this, we measured the period and
phase of collective oscillations in an array of spots spanning an area
of 9 × 9 mm2 (Fig. 2b) (Methods). Remarkably, the period remains
nearly identical across such a macroscopic area (Fig. 2c), while the
phase varies linearly over space (Fig. 2d). The collective oscillation of
cells is thus organized over centimetres (more than 104 times the cell
length) in the form of longitudinal travelling waves. In the example

a b
y long-axis orientation and the chirality were uncorrelated to the original
x
c already displaying collective oscillation (Methods; Extended Data
20
10
v (μm s–1)
0
–10
–20
0 5 10 15 20 25
Time (s)
Figure 1 | Collective oscillation in dense suspensions of E. coli.
a, Representative velocity field of cells’ collective motion obtained by PIV
analysis (see Methods). The red arrow indicates the direction of average
velocity. In all experimental figures throughout the paper, the +x axis
represents the colony expansion direction (Extended Data Fig. 1). b, In
the same field of view as a, two silicone oil tracers (red dots) displayed
synchronized oscillation in elliptical trajectories (Supplementary Video 2).
The background is the first frame of Supplementary Video 2. Scale bars in
a and b are 20 µm. c, Cells’ collective velocity (blue, x-axis component; red,
y-axis component) and tracer velocity (green, x-axis component; black,
y-axis component) in Supplementary Video 2 plotted against time.
given in Fig. 2d, the phase propagated with a wavelength of 17 mm and
a phase velocity of 3.7 mm s−1. Interestingly, the long axis of collective
oscillations tends to be perpendicular to the propagation direction of
travelling waves.
Next we examined individual trajectories of bacteria within colonies
displaying collective oscillation using fluorescent cells (Methods). As
suggested by phase-contrast videos (Supplementary Video 2), we found
that cells moved erratically without obvious periodic motion but with
frequent abrupt turns that are probably due to cell–cell collisions or
flagellar switching in a dense environment (Fig. 2e and Supplementary
Video 3). Cells also occasionally swam close to the surface and their
motion was then weakly biased towards the chirality of the collective
oscillation (Extended Data Fig. 4). Sharp turns among different cells
were not synchronized, and the time interval between two consecutive
turns of the same cell approximately followed an exponential distribu-
tion with an average of 2.1 ± 1.9 s (mean ± s.d., n = 118) (Extended Data
Fig. 5). This suggests that individual cells do not behave as oscillators,
and that the collective oscillation is an emergent weak synchronization
phenomenon at the population level, in contrast to most other collec-
tive biological oscillations studied2–11.
We then sought to control the emergence of the collective oscillations.
We cooled down entire oscillating colonies to a temperature low
enough to suppress any motion of the cells. Coming back then to the
original temperature, cells started moving again with normal speed and
in random directions (Methods). The collective oscillation typically
re-emerged from random motion in about 30 min. At onset an irregular
oscillation first emerged spontaneously, then its amplitude increased
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
LETTER RESEARCH
and a typical elliptic trajectory was recovered within less than one
minute (Fig. 3a, c and Extended Data Fig. 6a, b). Remarkably, the
one. These results confirm that collective oscillation is the result of
spontaneous symmetry breaking. In fact, we also found that collective
oscillation in cooled-off colonies may re-emerge simultaneously in
several regions with different chirality and we observed that the inter-
face separating two adjacent regions with collective oscillations of
opposite chirality moves, resulting in local chirality switching accom-
panied by a gradual change of the orientation of the ellipses (Fig. 3b, d
and Extended Data Fig. 6c, d). This observation suggests that travelling
waves seen in naturally developed colonies may result from the com-
petition of collective oscillations having emerged in different domains
and are not strongly coupled with colony development.
Finally, we investigated the mechanism driving collective oscillations.
We optically deactivated the motility of cells in a small area of colonies
Fig. 7 and Supplementary Video 4). We found that as the speed of the
cells decreased, the amplitude of collective oscillation, as measured by
the collective cell velocity and by tracer velocity, decreased as well and
eventually almost vanished when all cells became totally immotile, with
a small-amplitude residual oscillation reflecting the incompressibility of
fluids (Extended Data Fig. 7b). On the other hand, the collective oscilla-
tion remained unchanged beyond about 50 µm outside the boundary of
the motility-deactivated area (Extended Data Fig. 7c, d). These results
confirm that cell motility provides the driving force of collective
30
oscillation, and that the mechanism maintaining this emergent
phenomenon is local in nature and highly robust to perturbation.
We now present a mathematical model at the individual, ‘particle’
level that accounts for all our experimental observations. (We note that
our model cannot be fully conclusive at this stage because it is based on
assumptions that will need to be confirmed by a detailed study of the
complex interactions between bacteria bodies, flagella, gel substrate,
and the surrounding fluid.) First, in a minimalist approach, we consider
identical self-propelled particles with strictly local interactions, without
modelling the surrounding fluid explicitly. Second, to account for the
experimental observations left unexplained, we ‘immerse’ this model
in some Stokesian incompressible fluid.
In the Vicsek-style16,29, ‘dry’ core of our model, point particles move
at constant speed v0 in a two-dimensional domain without any repulsive
or attractive interaction. This allows us (1) to account for the fact that
in the experimental swarming fluid layer cells can pass above each
other, and (2) to simulate easily millions of cells, as required by the
large-scale, high-density context of the phenomena observed. Single-
particle dynamics involves an evolution equation for the velocity
direction θ that allows for local alignment, but also one for the angular
velocity or instantaneous frequency ω = θɺ to account for short-time
memory effects introduced by the local fluid vorticity and to allow for
synchronization of local rotational motion. Both these equations are
stochastic, with strong noise guaranteeing erratic individual
trajectories. Particles within a distance d0 of the order of the bacteria
body length are subjected to two interactions, a diffusive coupling
between angular velocities of strength kω, and a polar alignment of
strength kθ:
k
θɺi = ω i + θ ∑ sin(θj − θi ) + ξ θ (1)
ni j ~i
ωi k ω
ωɺ i = −
τ
+
ni
∑ (ω j − ω i ) + ξ ω + ξ bias (2)
j ~i
where the sums are over the ni particles j currently in the neighbour-
hood of i, and the delta-correlated noises ξθ and ξω are drawn from
symmetric uniform distributions in the intervals [ − η θ, η θ] and
[ − η ω , η ω ]. On the other hand, ξbias has the (current) sign of ω and its
9 F E B R UA RY 2 0 1 7 | VO L 5 4 2 | N AT U R E | 2 1 1
 RESEARCH LETTER

a 25 b c
16 1
20 4.7

Period (s)
15

N
4.5
10
5 4.3
y 0 4 8 12 16
0
0 30 60 90 120 150 x Spot
3 mm
Angle (°)
d e
1.6π
8
1.4π
1.2π
6
1.0π
y (mm)

0.8π
4
0.6π
0.4π
2
0.2π
0
0
0 2 4 6 8
x (mm)
Figure 2 | Collective oscillation organized into a centimetre-scale as ϕ = 0.25x + 0.27y (Methods), propagating at 0.37 rad mm−1 along its
travelling wave. a, Distribution of long-axis orientation of collective gradient direction. The double-ended red arrow indicates the long axis
oscillations (that is, the angle defined in Extended Data Fig. 1b) across of collective oscillation. e, Single-cell trajectories in the mid-layer during
N = 71 colonies. b, Sequence of measuring a 4 × 4 array of spots on a collective oscillation (crosses, starting points; pluses, ending points; see
colony undergoing collective oscillation (Methods). c, The period of Supplementary Video 3). An immotile cellular cluster (red trajectory) can
collective oscillation at all spots plotted in the order of measurement. serve as a flow tracer. Scale bar, 20 µm. The background is the last frame of
d, Contour map of the phase distribution of collective oscillation across Supplementary Video 3.
the same colony as in c. The phase (colour scale, in radians) can be fitted

a c
150 20
40

30 10
Velocity (μm s–1)

100
Time (s)
y (μm)

20
0
10
50
0 –10

–10
0 –20
–30 –20 –10 0 0 50 100 150
x (μm) Time (s)
b d
40 10

150
30 5
Velocity (μm s–1)
Time (s)
y (μm)

20 100
0

10
50 –5

0
0 –10
–30 –20 –10 0 0 50 100 150 200
x (μm) Time (s)
Figure 3 | Emergence of collective oscillation and chirality switching. in a and b were built from the collective cellular velocity obtained by PIV
a, Collective trajectory of cells during the emergence of collective analysis (the colour scale shows time in seconds). c, d, Collective velocities
oscillation. b, Chirality switching of collective oscillation from associated with a and b fitted by the smoothing spline method based on
counterclockwise to clockwise during domain competition. Trajectories PIV data (red, x-axis component; black, y-axis component) (see Methods).

2 1 2 | N AT U R E | VO L 5 4 2 | 9 F E B R UA RY 2 0 1 7
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 LETTER RESEARCH

a b c d e

Global polarity, P Global frequency, Ω

100 μm 1,200
0.8
0.6
0.4
0.2
0 900
0 50 100 150 200 250
Time (s)
0.4

y (μm)
0.3
600
0.2
0.1 16 2π
0
0 50 100 150 200 250 12 3π/2
Time (s)
12 300

Vx, Vy (μm s–1)

Vx 8 π
6 Vy
0
20 μm 4 π/2
–6
–12 0
0 50 100 150 0 0 5 10 15 20 0
Time (s) Time (s)
Figure 4 | Modelling results. a, Snapshot of ‘dry’ self-propelled particle wave regime, with particles coloured by their orientation θ (angular
model in a homogeneous, counterclockwise collective oscillation regime colour scale on the right of e) (Supplementary Video 6). The inset shows
(Supplementary Video 5). Trajectories of randomly selected particles the zoom of the elliptical trajectory built from the averaged velocity of
were shown during two periods of the collective oscillation cycle (start, particles in a 96 µm × 96 µm local domain. d, Fluid flow corresponding to c
blue; end, red). The line at the bottom left is a collective trajectory built (Supplementary Video 7). The colour scale shows the intensity of the local
from the averaged velocity of all particles (about 20,000). b, Time series of fluid speed (in micrometres per second). e, Spatiotemporal dynamics of
global frequency Ω, global polarity P, and collective velocity components the oscillation phase of Vx(y, t) showing the propagation of the travelling
Vx (horizontal) and Vy (vertical) from random initial conditions (same wave (angular colour scale).
parameters as in a). c, Snapshot of the ‘wet’ model in a stationary travelling

amplitude decreases with ω : ξ bias = sign(ω i )exp(− ω i /ω 0 )ξ where ξ trajectories, a type of ordered active matter not seen before, to our
is a uniform noise in [0, η bias]. This ‘bias’ noise, together with the knowledge. The phenomenon may influence spatial patterning of
diffusive coupling between angular velocities, gives the possibility of biofilms in bacterial colonies (Extended Data Fig. 9). This mechanism
chiral symmetry-breaking leading to a spatially homogeneous and of collective oscillation may be relevant to diverse biological processes
globally oscillating velocity field. Simulations of equations (1) to (2) that involve a large population of locally interacting cells with noisy and
showed that there is a large region of parameter space where collective non-oscillating individual behaviour.
oscillatory motion emerges: global angular velocity Ω = ω and global
polarity P = exp(iθ ) take finite values while individual trajectories Online Content Methods, along with any additional Extended Data display items and
Source Data, are available in the online version of the paper; references unique to
remain erratic (Fig. 4a, b; Methods). Not surprisingly for a diffusive these sections appear only in the online paper.
oscillatory medium30, travelling-wave configurations with arbitrary
large wavelength are stable. However, as indicated by the constancy Received 12 June; accepted 1 November 2016.
over time of Ω and P, average trajectories remain roughly circular in Published online 23 January 2017.
all cases (Fig. 4a, b), in contrast to experimental observations.
We now take into account the surrounding viscous fluid in a simpli- 1. Winfree, A. T. The Geometry of Biological Time (Springer, 1980).
fied way (see Methods for a more complete presentation). Cells interact 2. Oates, A. C., Morelli, L. G. & Ares, S. Patterning embryos with oscillations:
with the surface of the agar gel, pushing themselves away from it and structure, function and dynamics of the vertebrate segmentation clock.
entraining some fluid with them. The incompressible, damped fluid is Development 139, 625–639 (2012).
3. He, L., Wang, X., Tang, H. L. & Montell, D. J. Tissue elongation requires
governed by a Stokes equation, creating a flow field v that advects oscillating contractions of a basal actomyosin network. Nat. Cell Biol. 12,
particles: ɺr i = v + v0û i where û i is the unit vector of orientation θi. 1133–1142 (2010).
Simulations of this suspension model show that, in travelling-wave 4. Herzog, E. D. Neurons and networks in daily rhythms. Nat. Rev. Neurosci. 8,
790–802 (2007).
configurations, average trajectories of particles are ellipses with the long 5. Muller, P. et al. Differential diffusivity of Nodal and Lefty underlies a reaction-
axis perpendicular to the wave propagation direction, in agreement diffusion patterning system. Science 336, 721–724 (2012).
with experimental observations (Fig. 4c–e and Fig. 2d). The model also 6. Gregor, T., Fujimoto, K., Masaki, N. & Sawai, S. The onset of collective behavior
in social amoebae. Science 328, 1021–1025 (2010).
accounts well for the residual collective oscillations seen in motility- 7. Danino, T., Mondragon-Palomino, O., Tsimring, L. & Hasty, J. A synchronized
deactivation experiments (Extended Data Fig. 8 and Supplementary quorum of genetic clocks. Nature 463, 326–330 (2010).
Video 8). 8. Liu, J. et al. Metabolic co-dependence gives rise to collective oscillations within
biofilms. Nature 523, 550–554 (2015).
Taking all results together, we conclude that the emergence of collective 9. De Monte, S. et al. Dynamical quorum sensing: population density encoded
oscillatory motion in swarming flagellated bacteria is a robust self- in cellular dynamics. Proc. Natl Acad. Sci. USA 104, 18377–18381 (2007).
organized process probably mediated by local interactions. A defining 10. Bennett, M. V. L. & Zukin, R. S. Electrical coupling and neuronal
synchronization in the mammalian brain. Neuron 41, 495–511 (2004).
feature of this process is that individual cells remain strongly erratic 11. Koride, S. et al. Mechanochemical regulation of oscillatory follicle cell dynamics
while global order emerges. In contrast to other collective biological in the developing Drosophila egg chamber. Mol. Biol. Cell 25, 3709–3716
oscillations studied so far2–11, which mostly arise from phase-coupling (2014).
between explicit local oscillators, the phenomenon we report is 12. Ramaswamy, S. The mechanics and statistics of active matter. Annu. Rev.
Condens. Matter Phys. 1, 323–345 (2010).
characterized by two key factors. First, our system does not have 13. Marchetti, M. C. et al. Hydrodynamics of soft active matter. Rev. Mod. Phys. 85,
‘obvious’ local oscillators—they emerge only from noise upon sufficient 1143–1189 (2013).
coarse-graining; second, our synchronization is of erratic trajectories, 14. Nédélec, F. J., Surrey, T., Maggs, A. C. & Leibler, S. Self-organization of
microtubules and motors. Nature 389, 305–308 (1997).
that is, it occurs in physical space but not in phase space. As such, 15. Schaller, V., Weber, C., Semmrich, C., Frey, E. & Bausch, A. R. Polar patterns of
this phenomenon may constitute weak synchronization of stochastic driven filaments. Nature 467, 73–77 (2010).

9 F E B R UA RY 2 0 1 7 | VO L 5 4 2 | N AT U R E | 2 1 3
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
 RESEARCH LETTER
16. Sumino, Y. et al. Large-scale vortex lattice emerging from collectively moving
microtubules. Nature 483, 448–452 (2012).
17. Sanchez, T., Chen, D. T. N., DeCamp, S. J., Heymann, M. & Dogic, Z.
Spontaneous motion in hierarchically assembled active matter. Nature 491,
431–434 (2012).
18. Bricard, A., Caussin, J.-B., Desreumaux, N., Dauchot, O. & Bartolo, D.
Emergence of macroscopic directed motion in populations of motile colloids.
Nature 503, 95–98 (2013).
19. Rubenstein, M., Cornejo, A. & Nagpal, R. Programmable self-assembly in a
thousand-robot swarm. Science 345, 795–799 (2014).
20. Kearns, D. B. A field guide to bacterial swarming motility. Nat. Rev. Microbiol. 8,
634–644 (2010).
21. Darnton, N. C., Turner, L., Rojevsky, S. & Berg, H. C. Dynamics of bacterial
swarming. Biophys. J. 98, 2082–2090 (2010).
22. Wu, Y. & Berg, H. C. A water reservoir maintained by cell growth fuels the
spreading of a bacterial swarm. Proc. Natl Acad. Sci. USA 109, 4128–4133
(2012).
23. Zhang, H. P., Be’er, A., Smith, R. S., Florin, E.-L. & Swinney, H. L.
Swarming dynamics in bacterial colonies. Europhys. Lett. 87, 48011
(2009).
24. Sokolov, A. & Aranson, I. S. Physical properties of collective motion in
suspensions of bacteria. Phys. Rev. Lett. 109, 248109 (2012).
25. Wensink, H. H. et al. Meso-scale turbulence in living fluids. Proc. Natl Acad.
Sci. USA 109, 14308–14313 (2012).
26. López, H. M., Gachelin, J., Douarche, C., Auradou, H. & Clément, E. Turning
bacteria suspensions into superfluids. Phys. Rev. Lett. 115, 028301
(2015).
27. Lushi, E., Wioland, H. & Goldstein, R. E. Fluid flows created by swimming
bacteria drive self-organization in confined suspensions. Proc. Natl Acad.
Sci. USA 111, 9733–9738 (2014).
28. Pushkin, D. O., Shum, H. & Yeomans, J. M. Fluid transport by individual
microswimmers. J. Fluid Mech. 726, 5–25 (2013).
2 1 4 | N AT U R E | VO L 5 4 2 | 9 F E B R UA RY 2 0 1 7
© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
29. Vicsek, T. et al. Novel type phase transition in a system of self-driven particles.
Phys. Rev. Lett. 75, 1226–1229 (1995).
30. Kuramoto, Y. Chemical Oscillations, Waves, and Turbulence (Springer, 1984).
Supplementary Information is available in the online version of the paper.
Acknowledgements We thank Y. Li and W. Zuo from our laboratory for building
the microscope stage temperature control system, L. Xu (The Chinese University
of Hong Kong) for providing silicone oil, H. C. Berg (Harvard University) for
providing the bacterial strains, and J. Näsvall and J. Bergman (Uppsala
University) for providing the mRFP plasmid. This work was supported by the
Research Grants Council of Hong Kong SAR (RGC numbers 2191031 and
2130439 and CUHK Direct Grant numbers 3132738 and 3132739 to Y.W.),
the National Natural Science Foundation of China (NSFC number 21473152,
to Y.W.; NSFC number 11635002 to H.C. and X.S.; NSFC numbers 11474210,
11674236 and 91427302 to X.S.), and the Agence Nationale de la Recherche
(project Bactterns, to H.C.).
Author Contributions Y.W. discovered the phenomenon and designed the study.
C.C. and S.L. performed experiments. C.C., S.L., Y.W. analysed and interpreted
the data, with input from H.C. X.S. and H.C. developed the mathematical model.
All authors wrote the paper.
Author Information Reprints and permissions information is available
at www.nature.com/reprints. The authors declare no competing
financial interests. Readers are welcome to comment on the online version
of the paper. Correspondence and requests for materials should be
addressed to Y.W. (ylwu@phy.cuhk.edu.hk), H.C. (hugues.chate@cea.fr) or
X.S. (xqshi@suda.edu.cn).
Reviewer Information Nature thanks J. Hasty and the other anonymous
reviewer(s) for their contribution to the peer review of this work.